rfc6902

rfc6902

Complete implementation of RFC6902 "JavaScript Object Notation (JSON) Patch" (including RFC6901 "JavaScript Object Notation (JSON) Pointer"), for creating and consuming application/json-patch+json documents. Also offers "diff" functionality without using Object.observe.

Demo

Simple web app using the browser-compiled version of the code.

Quickstart

Install locally

npm install --save rfc6902

Import in your script

const rfc6902 = require('rfc6902')

Calculate diff between two objects

rfc6902.createPatch({first: 'Chris'}, {first: 'Chris', last: 'Brown'})
//⇒ [ { op: 'add', path: '/last', value: 'Brown' } ]

Apply a patch to some object

const users = [{first: 'Chris', last: 'Brown', age: 20}]
rfc6902.applyPatch(users, [
  {op: 'replace', path: '/0/age', value: 21},
  {op: 'add', path: '/-', value: {first: 'Raphael', age: 37}},
])

The applyPatch function returns [null, null], indicating there were two patches, both applied successfully.

The users variable is modified in place; evaluate it to examine the end result:

users
//⇒ [ { first: 'Chris', last: 'Brown', age: 21 },
//    { first: 'Raphael', age: 37 } ]

API

In ES6 syntax:

import {applyPatch, createPatch} from 'rfc6902'

Using TypeScript annotations for clarity:

applyPatch(object: any, patch: Operation[]): Array<Error | null>

The operations in patch are applied to object in-place. Returns a list of results as long as the given patch. If all operations were successful, each item in the returned list will be null. If any of them failed, the corresponding item in the returned list will be an Error instance with descriptive .name and .message properties.

createPatch(input: any, output: any, diff?: VoidableDiff): Operation[]

Returns a list of operations (a JSON Patch) of the required operations to make input equal to output. In most cases, there is more than one way to transform an object into another. This method is more efficient than wholesale replacement, but does not always provide the optimal list of patches. It uses a simple Levenshtein-type implementation with Arrays, but it doesn't try for anything much smarter than that, so it's limited to remove, add, and replace operations.

Optional diff argument

The optional diff argument allows the user to specify a partial function that's called before the built-in diffAny function. For example, to avoid recursing into instances of a custom class, say, MyObject:

function myDiff(input: any, output: any, ptr: Pointer) {
  if ((input instanceof MyObject || output instanceof MyObject) && input != output) {
    return [{op: 'replace', path: ptr.toString(), value: output}]
  }
}
const my_patch = createPatch(input, output, myDiff)

This will short-circuit on encountering an instance of MyObject, but otherwise recurse as usual.

Operation

interface Operation {
  op: 'add' | 'remove' | 'replace' | 'move' | 'copy' | 'test'
  from?: string
  path?: string
  value?: string
}

Different operations use different combinations of from / value; see JSON Patch (RFC6902) below.

Changelog

I'm not going to copy & paste my relatively descriptive commit messages into groups here; rather, these are just the changes that merited major version bumps:

4.x.x → 5.0.0 (2021-12-15)

• Short-circuits JSON pointer traversal over the prototype-polluting tokens __proto__, constructor, and prototype. I.e., /a/__proto__/b and /a/b evaluate to the same thing. This is in violation of the spec, which makes no special provisions for this idiosyncrasy of the JavaScript language, but AFAIK there's no way to strictly comply with the spec in JavaScript. It would probably be more correct to throw an error in those cases, but this 'solution' feels less disruptive / more in line with workarounds implemented by other libraries.

3.x.x → 4.0.0 (2020-07-27)

• Potential performance regression due to consolidating separate compare(a, b): boolean and diff(a, b): Operation[] logic into basically defining compare(a, b) as !diff(a, b).length (i.e., diff(a, b) returns empty array). This simplifies the codebase and ensures underlying semantics do not diverge, but potentially does unnecessary work in computing a full diff when all we really care about is whether there is at least one difference. It also facilitates the user completely specifying custom diff functionality with just one diff function, as opposed to a diff and corresponding compare (and avoids the headache of having to propagate both of those around internally).

2.x.x → 3.0.0 (2018-09-17)

• Corrects improper behavior in a few buggy edge cases, which might conceivably break consumers relying on incorrect behavior. (Tbh that applies to most bugfixes but I felt there were enough to add up to incrementing the major version.)
• Also moves around some of the internal API that was not intended to be used externally, but technically was exported. If you're only importing the public API of applyPatch, createPatch, and createTests from 'rfc6902', nothing has changed.

Implementation details

Determinism

If you've ever implemented Levenshtein's algorithm, or played tricks with git rebase to get a reasonable sequence of commits, you'll realize that computing diffs is rarely deterministic. E.g., to transform the string ab → bc, you could:

1. Delete a (⇒ b)
2. and then append c (⇒ bc)

Or...

1. Replace b with c (⇒ ac)
2. and then replace a with b (⇒ bc)

Both consist of two operations, so either one is a valid solution.

Applying json-patch documents is much easier than generating them, which might explain why, when I started this project, there were more than five patch-applying RFC6902 implementations in NPM, but none for generating a patch from two distinct objects. (There was one that used Object.observe(), which only works when you're the one making the changes, and only as long as Object.observe() hasn't been deprecated, which it has.)

So when comparing your data objects, you'll want to ensure that the patches it generates meet your needs. The algorithm used by this library is not optimal, but it's more efficient than the strategy of wholesale replacing everything that's not an exact match.

Of course, this only applies to generating the patches. Applying them is deterministic and unambiguously specified by RFC6902.

JSON Pointer (RFC6901)

The RFC is a quick and easy read, but here's the gist:

• JSON Pointer is a system for pointing to some fragment of a JSON document.
• A pointer is a string that is composed of zero or more /reference-token parts.
  • When there are zero (the empty string), the pointer indicates the entire JSON document.
  • Otherwise, the parts are read from left to right, each one selecting part of the current document, and presenting only that fragment of the document to the next part.
• The reference-token bits are usually Object keys, but may also be (decimal) numerals, to indicate array indices.

E.g., consider the NPM registry:

{
  "_updated": 1417985649051,
  "flickr-with-uploads": {
    "name": "flickr-with-uploads",
    "description": "Flickr API with OAuth 1.0A and uploads",
    "repository": {
      "type": "git",
      "url": "git://github.com/chbrown/flickr-with-uploads.git"
    },
    "homepage": "https://github.com/chbrown/flickr-with-uploads",
    "keywords": [
      "flickr",
      "api",
      "backup"
    ],
    ...
  },
  ...
}

1. /_updated: this selects the value of that key, which is just a number: 1417985649051
2. /flickr-with-uploads: This selects the entire object:

   {
     "name": "flickr-with-uploads",
     "description": "Flickr API with OAuth 1.0A and uploads",
     "repository": {
       "type": "git",
       "url": "git://github.com/chbrown/flickr-with-uploads.git"
     },
     "homepage": "https://github.com/chbrown/flickr-with-uploads",
     "keywords": [
       "flickr",
       "api",
       "backup"
     ],
     ...
   }
   

3. /flickr-with-uploads/name: this effectively applies the /name pointer to the result of the previous item, which selects the string, "flickr-with-uploads".
4. /flickr-with-uploads/keywords/1: Array indices start at 0, so this selects the second item from the keywords array, namely, "api".

Rules:

• A pointer, if it is not empty, must always start with a slash; otherwise, it is an "Invalid pointer syntax" error.
• If a key within the JSON document contains a forward slash character (which is totally valid JSON, but not very nice), the / in the desired key should be replaced by the escape sequence, ~1.
• If a key within the JSON document contains a tilde (again valid JSON, but not very common), the ~ should be replaced by the other escape sequence, ~0. This allows keys containing the literal string ~1 (which is especially cruel) to be referenced by a JSON pointer (e.g., /~01 should return true when applied to the object {"~1":true}).
• All double quotation marks, reverse slashes, and control characters must escaped, since a JSON Pointer is a JSON string.
• A pointer that refers to a non-existent value counts as an error, too. But not necessarily as fatal as a syntax error.

Example

This project implements JSON Pointer functionality; e.g.:

const {Pointer} = require('rfc6902')
const repository = {
  contributors: ['chbrown', 'diachedelic', 'nathanrobinson', 'kbiedrzycki', 'stefanmaric']
}
const pointer = Pointer.fromJSON('/contributors/0')
//⇒ Pointer { tokens: [ '', 'contributors', '0' ] }
pointer.get(repository)
//⇒ 'chbrown'

JSON Patch (RFC6902)

The RFC is only 18 pages long, but here are the basics:

A JSON Patch document is a JSON document such that:

• The MIME Type is application/json-patch+json
• The file extension is .json-patch
• It is an array of patch objects, potentially empty.
• Each patch object has a key, op, with one of the following six values, and an operator-specific set of other keys.
  • add: Insert the given value at path. Or replace it, if it already exists. If the parent of the intended target does not exist, produce an error. If the final reference-token of path is "-", and the parent is an array, append value to it.
    • path: JSON Pointer
    • value: JSON object
  • remove: Remove the value at path. Produces an error if it does not exist. If path refers to an element within an array, splice it out so that subsequent elements fill in the gap (decrementing the length of the array).
    • path: JSON Pointer
  • replace: Replace the current value at path with value. It's exactly the same as performing a remove operation and then an add operation on the same path, since there must be a pre-existing value.
    • path: JSON Pointer
    • value: JSON object
  • move: Remove the value at from, and set path to that value. There must be a value at from, but not necessarily at path; it's the same as performing a remove operation, and then an add operation, but on different paths.
    • from: JSON Pointer
    • path: JSON Pointer
  • copy: Get the value at from and set path to that value. Same as move, but doesn't remove the original value.
    • from: JSON Pointer
    • path: JSON Pointer
  • test: Check that the value at path is equal to value. If it is not, the entire patch is considered to be a failure.
    • path: JSON Pointer
    • value: JSON object

License

Copyright 2014-2021 Christopher Brown. MIT Licensed.